Bob Satellite Package Development and Maintenance

This tutorial explains how to build and distribute Python-based working environments for Bob. By following these instructions you will be able to:

  • Create a basic working environment using either a stock Bob installation or your own compiled (and possibly uninstalled) version of Bob;
  • Install additional python packages to augment your virtual work environment capabilities - e.g., to include a new python package for a specific purpose covering functionality that does not necessarily exists in Bob or any available Satellite Package;
  • Distribute your work to others in a clean and organized manner.

These instructions heavily rely on the use of Python distutils and zc.buildout. One important advantage of using zc.buildout is that it does not require administrator privileges for setting up any of the above. Furthermore, you will be able to create distributable environments for each project you have. This is a great way to release code for laboratory exercises or for a particular publication that depends on Bob.

Note

The core of our strategy is based on standard tools for defining and deploying Python packages. If you are not familiar with Python’s setuptools, distutils or PyPI, it can be beneficial to learn about those before you start. Python Setuptools and Distutils are mechanisms to define and distribute python code in a packaged format, optionally through PyPI, a web-based Python package index and distribution portal.

zc.buildout is a tool to deploy Python packages locally, automatically setting up and encapsulating your work environment.

Anatomy of a buildout Python package

The best way to create your package is to download a skeleton from the Idiap github website and build on it, modifying what you need. Fire-up a shell window and than do this:

$ git clone --depth=1 https://github.com/idiap/bob.project.example.git
$ cd bob.project.example
$ rm -rf .git #this is optional - you won't need the .git directory

We now recommend you read the file README.rst situated at the root of the just downloaded material. It contains important information on other functionality such as document generation and unit testing, which will not be covered on this introductory material.

The anatomy of a minimal package should look like the following:

.
+-- MANIFEST.in   # extras to be installed, besides the python files
+-- README.rst    # a description of the package, in restructured-text format
+-- bootstrap.py  # stock script downloaded from zc.buildout's website
+-- buildout.cfg  # buildout configuration
+-- setup.py      # installation + requirements for this particular package
+-- docs          # documentation directory
|   +-- conf.py   # Sphinx configuration
|   +-- index.rst # Documentation starting point for Sphinx
+-- bob          # python package (a.k.a. "the code")
|   +-- example
|   |   +-- script
|   |   |   +-- __init__.py
|   |   |   +-- version.py
|   |   +-- __init__.py
|   |   +-- test.py
|   +-- __init__.py

Our example that you just downloaded contains these files and a few extra ones useful for this tutorial. Inspect the package so you are aware of its contents. All files are in text format and should be heavily commented. The most important file that requires your attention is setup.py. This file contains the basic information for the Python package you will be creating. It defines scripts the package provides and dependencies it requires for execution. To customize the package to your needs, you will need to edit this file and modify it accordingly. Before doing so, it is suggested you go through all of this tutorial so you are familiar with the whole environment. The example package, as it is distributed, contains a fully working example.

In the remainder of this document, we explain how to setup buildout.cfg so you can work in different operational modes - the ones which are more common development scenarios.

Pure-Python Packages

Pure-Python packages are the most common. They contain code that is exclusively written in Python. This contrasts to packages that are written in a mixture of Python and C/C++.

The package you cloned above is a pure-Python example package and contains all elements to get you started. It defines a single library inside called bob.example, which declares a simple script, called version.py that prints out the version of Bob. When you clone the package, you will not find any executable as buildout needs to check all dependencies and install missing ones before you can execute anything. Here is how to go from nothing to everything:

$ python bootstrap.py
Creating directory '/home/user/work/tmp/bob.project.example/bin'.
Creating directory '/home/user/work/tmp/bob.project.example/parts'.
Creating directory '/home/user/work/tmp/bob.project.example/eggs'.
Creating directory '/home/user/work/tmp/bob.project.example/develop-eggs'.
Generated script '/home/user/work/tmp/bob.project.example/bin/buildout'.
$ ./bin/buildout
Develop: '/remote/filer.gx/user.active/aanjos/work/tmp/bob.project.example/.'
Getting distribution for 'bob.buildout'.
Got bob.buildout 0.2.13.
Getting distribution for 'zc.recipe.egg>=2.0.0a3'.
Got zc.recipe.egg 2.0.0.
Installing scripts.
...

Note

The Python shell used in the first line of the previous command set determines the Python interpreter that will be used for all scripts developed inside this package. To build your environment around a different version of Python, just make sure to correctly choose the interpreter you wish to use. If you just want to get things rolling, using python bootstrap.py will, in most cases, do the right thing.

You should now be able to execute ./bin/version.py:

$ ./bin/version.py
The installed version of bob.blitz is `2.0.0a0'
bob.blitz is installed at `...'
bob.blitz depends on the following Python packages:
 * bob.extension: 0.3.0a0 (...)
 * numpy: 1.6.2 (/usr/lib/python2.7/dist-packages)
 * distribute: 0.6.28dev-r0 (/usr/lib/python2.7/dist-packages)
 * coverage: 3.7.1 (...)
 * sphinx: 1.1.3 (/usr/lib/python2.7/dist-packages)
 * nose: 1.1.2 (/usr/lib/python2.7/dist-packages)
 * docutils: 0.8.1 (/usr/lib/python2.7/dist-packages)
 * jinja2: 2.6 (/usr/lib/python2.7/dist-packages)
 * pygments: 1.5 (/usr/lib/python2.7/dist-packages)
bob.blitz depends on the following C/C++ APIs:
 * Python: 2.7.3
 * Boost: 1.50.0
 * Blitz++: 0.10
 * NumPy: 0x01000009
 * Compiler: ('gcc', '4.7.2')

Everything is now setup for you to continue the development of this package. Modify all required files to setup your own package name, description and dependencies. Start adding files to your library (or libraries) and, if you wish, make this package available in a place with public access to make your research public. We recommend using Github. Optionally, drop-us a message talking about the availability of this package so we can add it to the growing list of Satellite Packages.

C or C++/Python Packages

Creating C++/Python bindings should be rather. Firstly, edit your setup.py so that you include the following:

from setuptools import setup, find_packages, dist
dist.Distribution(dict(setup_requires=['bob.blitz']))
from bob.blitz.extension import Extension
...

setup(

  name="bob.myext",
  version="1.0.0",
  ...
  install_requires=[
    'setuptools',
    'bob.blitz',
  ],
  ...
  namespace_packages=[
    'bob',
  ],
  ...
  ext_modules=[
    Extension("bob.myext._myext",
      [
        "bob/myext/ext/file1.cpp",
        "bob/myext/ext/file2.cpp",
        "bob/myext/ext/main.cpp",
      ],
      packages = [ #other c/c++ api dependences
        'bob-math',
        'bob-sp',
        ]
      ),
    ... #add more extensions if you wish
  ],

  ...
  )

These modifications will allow you to compile extensions that are linked against our core Python-C++ bridge bob.blitz. You can specify any pkg-config module and that will be linked in (for example, bob-ip or opencv) using the packages setting as shown above. Other modules and options can be set manually using the standard options for python extensions.

Documenting your C/C++ Python Extension

One part of this package are some functions that makes it easy to generate a proper python documentation for your bound C/C++ functions. This documentation can be used after:

#include <bob.extension/documentation.h>

Function documentation

To generate a properly aligned function documentation, you can use:

static bob::extension::FunctionDoc description(
  "function_name",
  "Short function description",
  "Optional long function description"
);

Note

Please assure that you define this variable as static.

Note

If you want to document a member function of a class, you should use set fourth boolean option to true. This is required since the default python class member documentation is indented four more spaces, which we need to balance:

static bob::extension::FunctionDoc member_function_description(
  "function_name",
  "Short function description",
  "Optional long function description",
  true
);

Using this object, you can add several parts of the function that need documentation:

  1. description.add_prototype("variable1, variable2", "return1, return2"); can be used to add function definitions (i.e., ways how to use your function). This function needs to be called at least once. If the function does not define a return value, it can be left out (in which case the default "None" is used).
  2. description.add_parameter("variable1, variable2", "datatype", "Variable description"); should be defined for each variable that you have used in the prototypes.
  3. description.add_return("return1", "datatype", "Return value description"); should be defined for each return value that you have used in the prototypes.

Note

All these functions return a reference to the object, so that you can use them in line, e.g.:

static auto description = bob::extension::FunctionDoc(...)
  .add_prototype(...)
  .add_parameter(...)
  .add_return(...)
;

Finally, when binding you function, you can use:

  1. description.name() to get the name of the function
  2. description.doc() to get the aligned documentation of the function, properly indented and broken at 80 characters (by default). This call will check that all parameters and return values are documented, and add a .. todo directive if not.

Sphinx directives like .. note::, .. warning:: or .. math:: will be automatically detected and aligned, when they are used as one-line directive, e.g.:

"(more text)\n\n.. note:: This is a note\n\n(more text)"

Also, enumerations and listings (using the * character to define a list element) are handled automatically:

"(more text)\n\n* Point 1\n* Point 2\n\n(more text)"

Note

Please assure that directives are surrounded by double \n characters (see example above) so that they are put as paragraphs. Otherwise, they will not be displayed correctly.

Note

The .. todo:: directive seems not to like being broken at 80 characters. If you want to use .. todo::, please call, e.g., description.doc(10000) to avoid line breaking.

Note

To increase readability, you might want to split your documentation lines, e.g.:

"(more text)\n"
"\n"
"* Point 1\n"
"* Point 2\n"
"\n"
"(more text)"

Leading white-spaces in the documentation string are handled correctly, so you can use several layers of indentation.

Class documentation

To document a bound class, you can use the bob::extension::ClassDoc("class_name", "Short class description", "Optional long class description") function to align and wrap your documentation. Again, during binding you can use the functions description.name() and description.doc() as above.

Additionally, the class documentation has a function to add constructor definitions, which takes an bob::extension::FunctionDoc object. The shortest way to get a proper class documentation is:

static auto my_class_doc =
    bob::extension::ClassDoc("class_name", "Short description", "Long Description")
      .add_constructor(
        bob::extension::FunctionDoc("class_name", "Constructor Description")
         .add_prototype("param1", "")
         .add_parameter("param1", "type1", "Description of param1")
      )
;

Note

The second parameter "" in add_prototype prevents the output type (which otherwise defaults to "None") to be written.

Note

For constructor documentations, there is no need to declare them as member functions. This is done automatically for you.

Currently, the ClassDoc allows to highlight member functions or variables at the beginning of the class documentation. This highlighting is still under development and might not work as expected.

Possible speed issues

In order to speed up the loading time of the modules, you might want to reduce the amount of documentation that is generated (though I haven’t experienced any speed differences). For this purpose, just compile your bindings using the “-DBOB_SHORT_DOCSTRINGS” compiler option, e.g. by adding it to the setup.py as follows (see also above):

...
ext_modules=[
  Extension("bob.myext._myext",
    [
      ...
    ],
    ...
    define_macros = [('BOB_SHORT_DOCSTRINGS',1)],
    ),
],
...

or simply define an environment variable BOB_SHORT_DOCSTRINGS=1 before invoking buildout.

In any of these cases, only the short descriptions will be returned as the doc string.

Document Generation and Unit Testing

If you intend to distribute your newly created package, please consider carefully documenting and creating unit tests for your package. Documentation is a great starting point for users and unit tests can be used to check funcionality in unexpected circumstances such as variations in package versions.

Documentation

To write documentation, use the Sphinx Document Generator. A template has been setup for you under the docs directory. Get familiar with Sphinx and then unleash the writer in you.

Once you have edited both docs/conf.py and docs/index.rst you can run the document generator executing:

$ ./bin/sphinx-build docs sphinx
...

This example generates the output of the sphinx processing in the directory sphinx. You can find more options for sphinx-build using the -h flag:

$ ./bin/sphinx-build -h
...

Note

If the code you are distributing corresponds to the work described in a publication, don’t forget to mention it in your README.rst file.

Unit Tests

Writing unit tests is an important asset on code that needs to run in different platforms and a great way to make sure all is OK. Test units are run with nose. To run the test unitson your package:

$ ./bin/nosetests -v
test_version (bob.example.test.MyTests) ... ok

----------------------------------------------------------------------
Ran 1 test in 0.001s

OK

Creating Database Satellite Packages

Database satellite packages are special satellite packages that can hook-in Bob‘s database manager bob_dbmanage.py. Except for this detail, they should look exactly like a normal package.

To allow the database to be hooked to the bob_dbmanage.py you must implement a non-virtual python class that inherits from bob.db.driver.Interface. Your concrete implementation should then be described at the setup.py file with a special bob.db entry point:

# bob database declaration
'bob.db': [
  'replay = bob.db.replay.driver:Interface',
  ],

At present, there is no formal design guide for databases. Nevertheless, it is considered a good practice to follow the design of currently existing database satellite packages. This should ease migration in case of future changes.

Python Package Namespace

We like to make use of namespaces to define combined sets of functionality that go well together. Python package namespaces are explained in details here together with implementation details. Two basic namespaces are available when you are operating with Bob or add-ons, such as database access APIs (shipped separately): the bob namespace is reserved for utilities built and shiped with Bob. The namespace bob (as for external Bob packages) should be used for all other applications that are meant to be distributed and augment Bob‘s features.

The example package you downloaded creates package inside the bob namespace called example. Examine this example in details and understand how to distributed namespace’d packages in the URL above.

In particular, if you are creating a database access API, please consider putting all of your package contents inside the namespace bob.db.<package>, therefore declaring two namespaces: bob and bob.db. All standard database access APIs follow this strategy. Just look at our currently existing database satellite packages for examples.

Distributing Your Work

To distribute a package, we recommend you use PyPI. The The Hitchhiker’s Guide to Packaging contains details and good examples on how to achieve this.

Version Numbering Scheme

We recommend you follow Bob‘s version numbering scheme using a 3-tier string: M.m.p. The value of M is a number starting at 1. This number is changed in case of a major release that brings new APIs and concepts to the table. The value of m is a number starting at 0 (zero). Every time a new API is available (but no conceptual modifications are done to the platform) that number is increased. Finally, the value of p represents the patch level, starting at 0 (zero). Every time we need to post a new version of Bob that does not bring incompatible API modifications, that number is increased. For example, version 1.0.0 is the first release of Bob. Version 1.0.1 would be the first patch release.

Note

The numbering scheme for your package and Bob‘s may look the same, but should be totally independent of each other. Bob may be on version 3.4.2 while your package, still compatible with that release could be on 1.4.5. You should state on your setup.py file which version of Bob your package is compatible with, using the standard notation defined for setuptools installation requirements for packages.

You may use version number extenders for alpha, beta, and candidate releases with the above scheme, by appending aN, bN or cN to the version number. The value of N should be an integer starting at zero. Python’s setuptools package will correctly classifier package versions following this simple scheme. For more information on package numbers, consult Python’s PEP 386. Here are lists of valid python version numbers following this scheme:

0.0.1
0.1.0a35
1.2.3b44
2.4.99c32

Release Methodology for Satellite Packages

Here is a set of steps we recommend you follow when releasing a new version of your satellite package:

  1. First decide on the new version number your package will get. If you are making a minor, API preserving, modification on an existing stable package (already published on PyPI), just increment the last digit on the version. Bigger changes may require that you signal them to users by changing the first digits of the package. Alpha, beta or candidate releases don’t need to have their main components of the version changed, just bump-up the last digit. For example 1.0.3a3 would become 1.0.3a4;

  2. In case you are making an API modification to your package, you should think if you would like to branch your repository at this position. You don’t have to care about this detail with new packages, naturally.

    If required, branching will allow you to still make modifications (patches) on the old version of the code and develop on the master branch for the new release, in parallel. It is important to branch when you break functionality on existing code - for example to reach compatibility with an upcoming version of Bob. After a few major releases, your repository should look somewhat like this:

    ----> time
    
    initial commit
    o---------------o---------o-----o-----------------------> master
                    |         |     |
                    |         |     |   v2.0.0
                    |         |     +---x----------> 2.0
                    |         |
                    |         | v1.1.0  v1.1.1
                    |         +-x-------x------> 1.1
                    |
                    |   v1.0.0  v1.0.1a0
                    +---x-------x-------> 1.0
    

    The o‘s mark the points in which you decided to branch your project. The x‘s mark places where you decided to release a new version of your satellite package on PyPI. The -‘s mark commits on your repository. Time flies from left to right.

    In this ficticious representation, the master branch continue under development, but one can see older branches don’t receive much attention anymore.

    Here is an example for creating a branch at github (many of our satellite packages are hosted there). Let’s create a branch called 1.1:

    $ git branch 1.1
    $ git checkout 1.1
    $ git push origin 1.1
    
  3. When you decide to release something publicly, we recommend you tag the version of the package on your repository, so you have a marker to what code you actually published on PyPI. Tagging on github would go like this:

    $ git tag v1.1.0
    $ git push && git push --tags
    

    Notice use prefix tag names with v.

  4. Finally, after branching and tagging, it is time for you to publish your new package on PyPI. When the package is ready and you have tested it, just do the following:

    $ python setup.py register #if you modified your setup.py or README.rst
    $ python setup.py sdist --formats=zip upload
    
    .. note::
      You can also check the .zip file that will be uploaded to PyPI before
      actually uploading it. Just call::
    
        $ python setup.py sdist --formats=zip upload
    
      and check what was put into the ``dist`` directory.
    
  5. Announce the update on the relevant channels.

Upload Additional Documentation to PythonHosted.org

In case you have written additional sphinx documentation in your satellite package that you want to share with the world, there is an easy way to push the documentation to PythonHosted.org. More detailed information are given here, which translates roughly into:

  1. Edit your setup.py and add the required package sphinx-pypi-upload:
setup(
  ...

  setup_requires=[
    ...
    'sphinx-pypi-upload',
  ],

  ...
)

And re-run buildout:

$ ./bin/buildout
  1. Create or edit the file setup.cfg in the root directory of your package. The content should be something like:
[build_sphinx]
source-dir = docs
build-dir  = build/sphinx
all_files  = 1

[upload_sphinx]
upload-dir = build/sphinx/html
  1. Create and upload the documentation:

    $ ./bin/python setup.py build_sphinx
    $ ./bin/python setup.py upload_sphinx
    

The link to the documentation will automatically be added to the PyPI page of your package. Usually it is a good idea to check the documentation after building and before uploading.

Satellite Packages Available

Look here for our growing list of Satellite Packages.